The Next Era of Games: Mods, Agents, and Infinite Worlds

1. Key Themes

Tool-Use Models Collapse the Idea-to-Prototype Loop

The core technical unlock is that frontier models can now directly operate professional game software rather than just advising on it. Jonathan Lai describes the full pipeline that one model can now run end-to-end. Anish Acharya adds that two separate model trends, one emergent and one deliberately trained, are converging.

Jonathan Lai: "In previous generations of models, they would just explain to the user, 'Oh, you should go and do this in Blender. You should do this in Unreal.' But Astra actually could like operate the software for you, and that was a huge unlock." 00:02:14

Anish Acharya: "These are two mega trends within models, one kind of emergent and one programmed, that are converging to drive a lot of the the acceleration in sort of game development." 00:05:45

AAA Development Cycles Strangle Innovation, and AI Can Fix the Risk Math

Lai argues the real cost of 10-year, hundred-million-dollar development cycles is not just money but conservatism: producers only greenlight safe bets. Cheaper cycles change what gets funded. Acharya adds the human cost of canceled projects and shifting taste.

Jonathan Lai: "If you're a producer at a AAA game studio, and you know it's going to take 10 years to ship a game and hundreds of millions of dollars, you're going to be very conservative in what you greenlight, right?" 00:06:33

Anish Acharya: "How many games were canceled 3 years into a 6-year cycle, or 5 years into a 6-year cycle." 00:06:05

Two Ways to Spend the Savings: Cheaper Games or Vastly Bigger Ones

Acharya frames a fork: use AI to ship the same games faster, or hold the timeline constant and increase ambition by an order of magnitude.

Anish Acharya: "The other actually way to explore this new capability, though, is to increase ambition. So imagine what the next 12-year game will look like. Like GTA VII." 00:07:43

Jonathan Lai: "AI can literally create an abundance era in games that wasn't previously possible. 'Cause we can make games that have never been seen before." 00:08:17

Who Can Make Games Is Changing, and Creation Becomes a Form of Consumption

Because games are multidisciplinary (code, art, design, sound), AI filling skill gaps expands the creator base dramatically. Acharya adds a distinctive thesis: most games will have an audience of one, the creator, and that is fine.

Jonathan Lai: "You no longer need to have a full team of people that are all like specialized in different roles, and instead, you can have AI kind of plug some of the gaps in your skills and knowledge." 00:09:12

Anish Acharya: "The vast majority of games are ever going to be played by one person, that is the creator, in the future. And that's okay." 00:12:30 He continues: "There's a sort of Zen gardening aspect to game creation, which will become increasingly popular." 00:12:46

Modding, Decompilation, and Remixing as the New Creation Primitive

Lai grounds this in history: multi-billion-dollar franchises began as mods. AI makes modding accessible to anyone, including an 8-year-old. Remixing also solves the blank-canvas problem.

Jonathan Lai: "Multi-billion dollar game franchises have been created using mods, and so this is just this is the history of somehow some of the best and and biggest games in the world have been created." 00:00:31

Jonathan Lai: "It's hard for someone to create something new using a blank canvas... Working within the framework of an existing game that someone else has made, and then just say, 'Hey, like add this thing, remix it with this other thing,' I think it becomes very powerful 'cause it gives you a place to start and a place to remix from." 00:19:43

From Dynamic Content to Dynamic Mechanics

Acharya describes a qualitative leap beyond procedural generation: coding agents that modify the game engine itself at runtime based on player prompts.

Anish Acharya: "In the past it was really inspirational to have procedural level generation, but that's just dynamic content. Now we're talking about like dynamic abilities. It's a whole different universe." 00:15:42

Dead and Niche Genres Become Economically Viable

Genres that never penciled out (RTS on tablets, job simulators, dating sims) become viable when production costs fall. Lai and Acharya both cite RTS specifically, and Lai cites Tim Sweeney making the same point.

Anish Acharya: "There's genres of games that have really high production cost or have historically had... where you simply don't get the ROI, like, you know, people love RTS games... now I think we're going to see this complete renaissance of genres like RTS." 00:22:34

Jonathan Lai: "You can basically role play a day in the life of almost any job in the world that you that you can envision, right? Like, wouldn't that be really cool?" 00:25:01

Human Taste Remains the Bottleneck: The Last 20%

The hosts probe whether AI can deliver "fun." Lai says taste and feel remain human territory for the foreseeable future.

Jonathan Lai: "There's a saying in the games industry, the last 20% of the game is 80% of the work. And that's where all the magic comes together to make something fun." 00:10:33

Jonathan Lai: "I think that is going to be the province of humans for the foreseeable future, right? It's like, how does something feel? What do players want?" 00:10:57

The Never-Ending, Adaptive Game as the Holy Grail

Lai describes the next wave beyond productivity: AI that generates quests, mechanics, and worlds in response to player desire, an MMO that is not scripted linearly.

Jonathan Lai: "The Holy Grail for an MMO is a game which infinitely sort of adapts itself to what the players want... You could literally generate an entire quest line for someone who really likes the ocean. And like now there's pirates that are attacking you." 00:35:58

Game Interfaces Become the UI for Agent Orchestration

The discussion of RTS-skinned agent managers suggests games may supply the interface paradigm for managing fleets of agents. Lai argues the mass market needs something more visual than a command line.

Anish Acharya: "Being the general manager of a line of business at any company... is a sort of sim... You have a limited amount of control, and you have incomplete, imperfect information, and then you have market forces." 00:26:18

Jonathan Lai: "The average consumer finds agent orchestration to be non-intuitive, right? Like working from a command line..." 00:26:53

Distribution and Business Models Are the Unsolved Gaps

IP owners historically capture little value from mods. A new AI-native distribution platform and revenue-sharing structures remain open.

Jonathan Lai: "Historically the original owners of IP have not benefited that much from the mods that have been created on top of their IP." 00:28:06

Jonathan Lai: "There is an opportunity for a company out there to also create a distribution platform which natively harnesses and empowers all these AI game creators that are making stuff." 00:30:00

2. Contrarian Perspectives

Most Games Will Be Played by Just One Person, and That's a Feature

The conventional view of games is as shared entertainment products optimized for audience size. Acharya argues the future majority of games are personal artifacts whose value is the act of making them.

Anish Acharya: "Sometimes creation is consumption. Like the vast majority of games are ever going to be played by one person, that is the creator, in the future. And that's okay, actually, 'cause I think that there is, you know, programming or creating games for the output, for the artifact. And then there's doing those things as an activity." 00:12:21

Supporting evidence: his own record-shop-turned-Queen-Street-1990s game, built for fun and then extended by his son.

Realism Is Not the Most Interesting Axis for Improving Games

The industry race has been toward fidelity (GTA's city detail, AAA graphics). Acharya argues novelty of concept combination matters more, citing an 8-year-old's Ghost of Tsushima Minecraft mod.

Anish Acharya: "It illustrates the fact that as we think about how games get better, realism is not the only, or even necessarily the most interesting, axis for improvement." 00:17:19

He adds: "Just the fact that you can create a concept from two existing concepts... that never would have existed, is extraordinary." 00:17:30

Forking an IP Hurts the Owner; Extending It Helps

Many treat "remix" and "extend" as interchangeable and assume AI modding is good for IP holders. Acharya draws a sharp line: additive content on a shared core builds the network, while forks split it.

Anish Acharya: "All of the additions that extend the core game and maybe share a core foundation of community plus a base game, that actually really benefits the IP owner. Whereas when you're forking the game, there is no benefit. In fact, you might even be splitting the community." 00:29:23

Niche Genres Have Huge Latent Demand Hidden by Production Economics

The assumption is that genres are small because demand is small. Lai argues the supply side suppressed discovery: people never found the job or sim they'd love because nobody could afford to build it.

Jonathan Lai: "Maybe there's like a particular type of job that resonates for various people, but like maybe you never had a chance to discover it, because the simulator genre just wasn't that big." 00:24:52

Anish adds that the sim format has deep historical pull: "If you look at like old computing magazines... number one is usually like VisiCalc and number two is like flight- like Microsoft Flight Simulator." 00:24:23 (spoken by one of the hosts)

AI Should Be Used to Make New Mechanics, Not Just GTA Faster and Cheaper

Most observers frame AI in games as a cost-reduction tool. Lai positions that as wave one and argues the real prize is new gameplay that expands the market.

Jonathan Lai: "You're not just making GTA VI, but like faster and cheaper. You're making a new type of game, right, that couldn't exist before." 00:35:48

3. Companies Identified

Rockstar Games / Grand Theft Auto VI. Studio behind GTA; GTA VI is the benchmark for long-cycle AAA development. Mentioned as the canonical example of the 12-13 year development problem and of obsessive detail. Jonathan Lai: "Grand Theft Auto VI is finally coming out next month. That game has been in production since like 2013, 2014. It's like a 12, 13-year development cycle." 00:00:00

Epic Games. Maker of Unreal Engine and Fortnite; CEO Tim Sweeney is publicly bullish on AI. Mentioned for supporting AI-enabled niche-genre development. Jonathan Lai: "That's something that Tim Sweeney talks about, actually, in some of his Twitter posts, which is that AI is now making it possible to build sort of niche games... that previously would not have been economically viable." 00:03:20

Xbox / Microsoft. Xbox CEO Asha is thinking about hardware and software interplay, with an "Xbox Labs" announcement. Lai links it to community projects decompiling Halo into a WebGPU game. Jonathan Lai: "There were so many community projects popping up that were taking Halo... and then they were decompiling and recompiling it for like modern sort of platforms." 00:21:04

Minecraft (Microsoft). Cited as the strongest example of an IP owner benefiting from a mods ecosystem and as the platform for the Ghost of Tsushima mod. Anish Acharya: "The strongest form in my mind is Minecraft, you know? Like Minecraft really benefits from the mods ecosystem." 00:29:11

Roblox. Decentralized creation platform with games like Adopt Me!, Jailbreak, and Brookhaven. Used as the contrast case to Minecraft's centralized model. Jonathan Lai: "Once you're inside the Roblox ecosystem, if you ever get bored of one game... there's like hundreds of thousands, if not millions more, sort of experiences that you can participate in." 00:31:08

Spawn (founder Jacob). Shared community plus game generation platform where an embodied agent named Savvy accompanies players, modifies games live, and creates portals between games. a16z is not an investor, but Acharya is a fan. Anish Acharya: "It's the most impressive, ambitious effort I've seen at doing something like this. And I'm sure there'll be more." 00:31:01 Lai confirms: "I'm a fan of Jacob and have been following the company for a while." 00:31:08

Anthropic (Claude / Opus). Model provider whose coding and tool-use models drove the shift. Acharya notes an unplanned capability: "It was awesome with Three.js, which is in-browser 3D... it's laying things out almost on an element-by-element basis. Like in a way that would be maddeningly meticulous for a human." 00:05:03 Also referenced via Claude Code instances rendered as structures in a Minecraft-style agent manager. 00:25:54

OpenAI (ChatGPT/Codex). Used by Lai's son to deconstruct Ghost of Tsushima and build a Minecraft mod. Jonathan Lai: "Essentially Codex helped him build a Minecraft mod, which he installed." 00:16:47

Valve (Half-Life / Counter-Strike). Cited as the modding precedent. Jonathan Lai: "Counter-Strike, for example, which is a really big esports, that was a Half-Life mod." 00:18:10

Riot Games (League of Legends). Multi-billion-dollar MOBA inspired by the Dota mod; also Lai's most-played game. Jonathan Lai: "League of Legends, one of the most successful MOBAs of all time, was inspired by Defense of the Ancients, Dota, which was itself a Warcraft III mod." 00:18:10

Blizzard (Warcraft III, StarCraft). Warcraft III hosted the Dota mod; StarCraft cited as a declining hardcore RTS. Jonathan Lai: "StarCraft... the audience for hardcore RTS has just dwindled over time, because it's such a hard genre to play." 00:38:32

Krafton (PUBG) and Epic (Fortnite). Battle royale genre descended from a PlayerUnknown mod of Arma. Jonathan Lai: "Fortnite and PUBG were itself inspired by a PlayerUnknown Battle Royale mod of Arma." 00:18:10

BioWare (Knights of the Old Republic). Out of business, so beloved titles are candidates for community forking. Anish Acharya: "BioWare is out of business and... is never going to make another Knights of the Old Republic game. And you think about extending it, or forking it, like that's really cool." 00:20:29

Habbo Hotel. Social virtual world that inspired Acharya's record-shop game. Anish Acharya: "I gave it just a couple Habbo Hotel reference art components and it super exceeded my expectations." 00:14:28

Unity and Unreal Engine. Engines that models can operate directly. Jonathan Lai: "It could actually work directly within Unreal and Unity." 00:02:14

Blender. 3D tool that models now operate for modeling, rigging, and animation. Anish Acharya: "Could use Blender and Blender in the cloud really, really effectively." 00:05:45

Three.js. In-browser 3D library that became a model strength. Jonathan Lai: "3D on the web has never been more powerful or more accessible." 00:22:10

Paradox Interactive (Crusader Kings III). Grand strategy game cited by a co-host as already approximating agent-driven rulership with personalities, schemes, and relationships. Co-host: "It actually feels the closest to me like a kind of never-ending and endless game." 00:42:31

Credit Karma and Google. Mentioned in passing as examples of line-of-business GM roles that resemble sims. Anish Acharya: "Being the general manager of a line of business at any company, at Credit Karma or Google or wherever else is it is a sort of sim." 00:26:18

Untitled Goose Game (House House). Mischief sim Acharya enjoyed, cited as evidence of sim-genre pull. 00:25:14

Witcher 3 (CD Projekt Red), Baldur's Gate 3, Call of Duty, Ghost of Tsushima, Animal Crossing, Civilization, Dune 2000, Age of Empires, Factorio, Ender's Game. Referenced as games or fiction anchoring taste: Witcher 3 (completed twice by Acharya), Baldur's Gate 3 (cited for depth that holds players for a thousand hours), Call of Duty ("the 20th installment" as sequel conservatism), and Dune 2000 (Acharya's favorite childhood RTS, ported to Mac). Anish Acharya: "Dune 2000. That was my favorite RTS growing up." 00:37:58

4. People Identified

Tim Sweeney (CEO, Epic Games). Publicly articulated that AI enables niche, esoteric genres. Jonathan Lai: "AI is now making it possible to build sort of niche games, or sort of like esoteric genres that previously would not have been economically viable." 00:00:46

Asha (CEO, Xbox). Spoke about hardware and software interplay and supporting AI development frontiers through Xbox Labs. Lai links the announcement to community decompilation of Halo. 00:21:04

Blendi (X creator). Posts demonstrations of combining assets from two existing retro games into a third. Anish Acharya: "He'll take the assets from two existing games, and then combine them into a third game... which is just a totally outer space concept, but it works and some of the games are really fun." 00:20:29

Jacob (founder, Spawn). Built the shared game-generation community with the embodied agent Savvy. Anish Acharya: "It's Jacob and team, and they've created this sort of shared community plus game generation platform." 00:30:24

Amit Patel (early Google employee). Wrote the BBS async multiplayer games Barren Realms Elite and Solar Realms Elite while at Stanford. Acharya recreated Solar Realms Elite, whose source was lost, from a design doc and an .exe. Anish Acharya: "He lost the source code for Solar Realms Elite, which was the better of the two, but he has a game design doc and he has an .exe... I now have what feels like a faithful recreation running on my computer." 00:40:39

Anish Acharya's son. Prompted a coding agent to create a pet adoption center with followable pets inside his father's game. Anish Acharya: "My son went in there and created this pet adoption center... Now of course I didn't design this like animal following capability in, but the coding agent updated my game engine to create that mechanic." 00:15:23

Jonathan Lai's 8-year-old son. Used Codex to build a Ghost of Tsushima Minecraft mod and now builds instead of requesting games. Jonathan Lai: "He doesn't ask us to buy new games anymore. He's just like, 'I want to play this game. I'm going to make it as a Minecraft mod.'" 00:17:10

Anish Acharya's DJ friend. Sparked the online record-shop idea by noting online stores all look the same. Anish Acharya: "He was saying, 'Oh, all the online record stores look the same. It's just a list of songs.'" 00:13:51

PlayerUnknown (Brendan Greene). Creator of the Arma battle royale mod that led to PUBG and Fortnite. 00:18:10

Orson Scott Card's Ender Wiggin (fictional). Used by Lai as the model for an adaptive "mind game." 00:37:05

5. Operating Insights

Use Existing IP as the Starting Canvas to Beat Blank-Prompt Paralysis

When building AI-assisted creation tools, give users a structure to modify rather than an empty prompt. Lai's insight generalizes beyond games: constraints and familiar references speed creation and raise the odds of a hit.

Jonathan Lai: "If you just put a text prompt in front of someone and say, 'Hey, like make me a game,' like, you wouldn't know where to start, right? ... It gives you a place to start and a place to remix from." 00:19:58

Let the Model QA Itself with Screenshots

A concrete workflow for any build-test loop: have the model capture its own output, critique it, and fix bugs, which is why early prototypes are surprisingly good.

Jonathan Lai: "Astra can also QA itself. And so it can take screen caps of the game, and then it can learn from its own output, 'Oh, I need to fix this bug,' or, 'This doesn't look right.'" 00:02:58

Build Extensibility into the Core So Users Can Prompt New Capabilities

Acharya's design choice, making storefronts that other players can build out through prompting including new mechanics, turned a single product into a platform where users created features the builder never designed.

Anish Acharya: "I made it so that every one of the other storefronts were something that another player could come and build out just through prompting, but the interesting thing is that they could not just customize the art, they could customize the capabilities." 00:15:07

Start from Social Experience, Not Catalog, When Reimagining a Category

The record shop project began by identifying what the offline experience had that the online one lacked (people, ambient music, conversations) and rebuilding that. The lens applies to commerce and community categories generally.

Anish Acharya: "Going to the record shop was like going to the barber shop... you can't not have five conversations." 00:13:51

Treat Cost Reduction as a Greenlight Strategy: More At-Bats on New IP and Unproven Teams

Studios and publishers should convert saved budget into portfolio breadth rather than only margin.

Jonathan Lai: "You can actually ship a new IP. You can take a bet on some game developers that have never actually shipped anything before and you can greenlight them and get them a budget to make something." 00:07:29

6. Overlooked Insights

Long-Lost Software Can Be Reconstructed from a Design Doc and a Binary

Acharya mentions almost in passing that a 1990s game whose source code was lost was faithfully recreated from just a design document and an executable, after earlier model generations failed. This signals a broader capability: reverse-engineering and reconstructing abandoned or orphaned software from artifacts. The implication reaches well beyond games to legacy enterprise systems, lost codebases, and abandoned products with a surviving spec or binary.

Anish Acharya: "He has a game design doc and he has an .exe. So I've, for every genre of the models, I've tried like, 'Hey, can you recreate the game?' And it has not been able to do it. And I like now have what feels like a faithful recreation running on my computer." 00:40:44

The Orphaned-IP Problem: Studios That Died Leave Value Locked Up

Acharya briefly notes BioWare is out of business and will never make another Knights of the Old Republic, while Lai notes new generations of kids have never played these games. Together they point to a latent asset class: dormant, culturally beloved IP with no active steward, now technically revivable by communities. Whoever builds the legal and commercial framework for licensing, royalty sharing, or stewarding these revivals could capture significant value, and Lai flags the missing business-model layer.

Anish Acharya: "BioWare is out of business and, you know, BioWare is never going to make another Knights of the Old Republic game." 00:20:29

Jonathan Lai: "You're basically taking an established IP and you're exposing it to a new audience. Like kids today that have never had a chance to play Knights of the Old Republic. But now they can." 00:22:11